making some past projects public
1. Multi-layer perception - word prediction given a sequence
2. CNN: is this a pair of shoes?
3. RNN, Autoencoder: news headlines generator.
- Currently just generates the same headline, but can be extended to have text as input, headline as output
- Audio analysis to extract frequencies and notes
- Above as the basis of a custom layer to train a model. Lots of problems integrating to tflite and react native.
- Theoretically chord detection can be done via math eg fourier transform. But my application is band music with multiple instruments, improvisations and jazzy overtones. Hence tried to combine mathematical feature extraction and ML, resulting in much headache.
- In retrospect this problem should not be solved with a GAN, but a regression model to predict the degree of distortion in 3 dimensions (vertical skew, horizontal skew, rotation) and transform the image representation
- Got some pretty cool & crazy architecture photos though